ArticlePDF Available

Temporal Characteristics of P-Band Tomographic Radar Backscatter of a Boreal Forest

January 2021
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing PP(99):1-1

DOI:10.1109/JSTARS.2021.3050611

License
CC BY 4.0

Authors:

Albert Monteith

Chalmers University of Technology

L.M.H. Ulander

Chalmers University of Technology

Temporal variations in synthetic aperture radar (SAR) backscatter over forests are of concern for any SAR mission with the goal of estimating forest parameters from SAR data. In this article, a densely sampled, two-year long time series of P-band (420 to 450 MHz) boreal forest backscatter, acquired by a tower-based radar, is analyzed. The experiment setup provides time series data at multiple polarizations. Tomographic capabilities allow the separation of backscatter at different heights within the forest. Temporal variations of these multi-polarized, tomographic radar observations are characterized and quantified. The mechanisms studied are seasonal variations, effects of freezing conditions, diurnal variations, effects of wind and the effects of rainfall on backscatter. An emphasis is placed on upper-canopy backscatter, which has been shown to be a robust proxy for forest biomass. The canopy backscatter was more stable than ground-level backscatter during non-frozen conditions, supporting forest parameter retrieval approaches based on tomography or interferometric ground notching. Large backscatter variations during frozen conditions, which may be detected using cross-polarised backscatter observations, can result in large errors in forest parameter estimates. Diurnal backscatter variations observed during hot periods were likely connected to tree water transport and storage mechanisms. Backscatter changes were also observed during strong winds. These variations were small in comparison to the variations due to freeze-thaw and soil moisture changes and should not result in significant forest parameter estimation errors. The presented results are useful for designing physically based and semi-empirical scattering models that account for temporal changes in scattering characteristics.

RGB aerial photo of the experiment site acquired on 10 May 2018 by the Swedish National Land Survey (Lantmäteriet). The bottom corner reflector was used for calibrating the tower radar and the top corner reflector was temporarily installed for airborne SAR observations.

…

Block diagram of the radar instrument.

…

Photo of the top section of the radar tower. The two antenna arrays were connected to the VNA on the ground via coaxial cables.

…

An example of the insertion loss and phase shift by a reference cable connected between a transmit-receive pair of VNA ports measured at 1-hour intervals at 435 MHz. There was very little variation over the duration of experiment, indicating that there was an insignificant amount of drift in the VNA's measurement response.

…

Tomographic image examples from 06:00 on 1 September 2018. Height profiles of the image backscatter within the regions surrounded by dotted lines are shown on the right of each tomogram. Backscatter peaks at the ground level and upper canopy level. The solid white line shows a lidarderived canopy height estimate, the brown line shows the ground level and the 50 m-high tower is illustrated to scale on the left in each tomogram.

…

Figures - available via license: Creative Commons Attribution 4.0 International

Content may be subject to copyright.

Available via license: CC BY 4.0

Content may be subject to copyright.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSTARS.2021.3050611, IEEE

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

uller199012354

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Temporal Characteristics of P-band Tomographic

Radar Backscatter of a Boreal Forest

Albert R. Monteith, Member, IEEE and Lars M. H. Ulander, Fellow, IEEE,

Abstract—Temporal variations in synthetic aperture radar

(SAR) backscatter over forests are of concern for any SAR

mission with the goal of estimating forest parameters from SAR

data. In this article, a densely sampled, two-year long time series

of P-band (420 to 450 MHz) boreal forest backscatter, acquired

by a tower-based radar, is analyzed. The experiment setup

provides time series data at multiple polarizations. Tomographic

capabilities allow the separation of backscatter at different

heights within the forest. Temporal variations of these multi-

polarized, tomographic radar observations are characterized

and quantiﬁed. The mechanisms studied are seasonal variations,

effects of freezing conditions, diurnal variations, effects of wind

and the effects of rainfall on backscatter. An emphasis is placed

on upper-canopy backscatter, which has been shown to be a

robust proxy for forest biomass. The canopy backscatter was

more stable than ground-level backscatter during non-frozen

conditions, supporting forest parameter retrieval approaches

based on tomography or interferometric ground notching. Large

backscatter variations during frozen conditions, which may

be detected using cross-polarised backscatter observations, can

result in large errors in forest parameter estimates. Diurnal

backscatter variations observed during hot periods were likely

connected to tree water transport and storage mechanisms.

Backscatter changes were also observed during strong winds.

These variations were small in comparison to the variations due

to freeze-thaw and soil moisture changes and should not result

in signiﬁcant forest parameter estimation errors. The presented

results are useful for designing physically based and semi-

empirical scattering models that account for temporal changes

in scattering characteristics.

Index Terms—Backscatter, boreal forest, P-band, time series.

I. INT ROD UC TI ON

FOREST backscatter measured using imaging radars varies

with time due to changes in weather conditions, partic-

ularly pronounced at higher latitudes due to strong seasonal

effects [1]–[4]. If such variations are not accounted for, they

may affect the accuracy of forest parameter estimates such as

forest height and biomass, as estimated from synthetic aperture

radar (SAR) data.

The European Space Agency’s BIOMASS SAR satellite is

scheduled for launch in 2023. The main scientiﬁc objective

of the mission is to quantify forest carbon stocks and ﬂuxes

through the mapping of above-ground biomass from SAR

This work was ﬁnancially supported by the Hildur and Sven Wingquist

Foundation for Forest Research, the European Space Agency (ESA) and the

Swedish National Space Agency.

Albert R. Monteith is with the Department of Space, Earth and Environ-

ment, Chalmers University of Technology, SE-412 96 Gothenburg, Sweden

(e-mail: albert.monteith@chalmers.se).

Lars. M. H. Ulander is with the Department of Space, Earth and Environ-

ment, Chalmers University of Technology, SE-412 96 Gothenburg, Sweden,

and also with the Radar Systems Unit, Swedish Defence Research Agency

(FOI), SE-581 11 Link¨

oping, Sweden (email: lars.ulander@chalmers.se).

observations [5], [6]. The BIOMASS SAR instrument will be

the ﬁrst ever spaceborne SAR operating at P-band (centered at

435 MHz). This relatively low frequency, compared to existing

spaceborne SARs, allows the emitted electromagnetic waves

to penetrate the canopy and reﬂect off larger structures such

as branches and tree stems where the majority of a tree’s

biomass is located. This makes P-band especially sensitive to

above-ground forest biomass [7]–[9]. A consequence of the

increased canopy penetration is that the ground contributes

signiﬁcantly to the total backscattered ﬁeld, either through

direct rough surface scattering or double-bounce scattering

by the ground and tree trunks [10]. Variations in ground

roughness, slope and soil moisture can thus introduce a

signiﬁcant bias in the estimated biomass [11]. The BIOMASS

mission is designed for fully-polarimetric, interferometric and

tomographic imaging. Due to P-band transmission regulations

over Europe and North America, all of the world’s forests will

not be covered by BIOMASS. The loss in terms of the global

forest above-ground biomass carbon stock is estimated to be

18.2%, whereas the corresponding value for boreal forests is

36.7% [12]. In boreal forests, the combination of polarimetric

channels in SAR observations have been shown to reduce the

inﬂuence of the ground [13]. In tropical forests, tomographic

intensity near 30 m above the ground was shown to be

less impacted by terrain topography and closely correlated

with forest biomass [14]. Tomographic ground separation has

also been achieved in boreal forests [15]. More recently,

a ground-cancelling technique was developed whereby the

ground contribution is suppressed by coherently combining

interferometric image pairs, isolating the above-ground canopy

contribution [16], [17]. Central to all these biomass estimation

approaches is the use of the measured backscatter, especially

that of the above-ground canopy, to estimate biomass. Even

with the ground component removed, the canopy backscatter

does not uniquely depend on biomass. Changes in canopy

backscatter due to weather and seasonal changes will affect the

estimated biomass. Very little is known about the characteris-

tics of such temporal variations, especially in boreal forests,

making it difﬁcult to design biomass estimation algorithms

that are robust to temporal variations. The lack of a quanti-

tative understanding of environmentally induced backscatter

variations is also the main issue in developing algorithms for

forest degradation detection [11].

Forest backscatter observed by a SAR is governed by

the geometry of forest structures (ground, stems, branches

and leaves/needles) and the relative permittivities of these

structures [18]. The geometry and permittivity determine the

angular distribution and strength of these reﬂections as well as

the absorption of electromagnetic energy in forest structures.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Backscatter variations occur due to changes in either the

geometry or permittivity of forest structures

Apart from tree growth, geometric changes occur mainly

due to wind-induced tree swaying. The resulting displacement

of scatterers can be close to, or larger than, the wavelength,

and thus will affect the scattered ﬁeld measured by a radar.

To a lesser extent, terrestrial lidar observations have shown

that geometric changes occur due to diurnal drooping and

rising of branches [19]. Geometric changes therefore occur

at timescales of seconds (wind) to years (tree growth).

For most cases, an increase in the permittivity of the soil

or vegetation increases the forest backscatter [20]. Changes in

the permittivity of forest structures have been observed to be

caused by a variety of mechanisms that affect the water content

and chemical composition of forest materials [21]. A higher

soil water content increases the permittivity of the soil and

results in a stronger ground reﬂection [22]–[24]. Mechanisms

driving water content variations in trees are signiﬁcantly more

complex. The most widely accepted theory of water transport

in trees is the cohesion-tension theory [25]. According to the

cohesion-tension theory, water moves from the roots to the

stomata (pores in leaves/needles for gas exchange with the

atmosphere) as continuous columns of water as water is drawn

out of stomata during transpiration. The ﬂow of water from

roots to stomata can be characterized by resistances, limiting

the rate of upward ﬂow, and capacitances, representing water

storages in the xylem (sapwood) and phloem (bark) [26], [27].

These water reserves are depleted in hot and dry conditions

when the rate of transpiration exceeds the rate of soil water

uptake, resulting in a change of permittivity of tree struc-

tures. The rate of transpiration is controlled by a complicated

interaction between solar radiation, air temperature, relative

humidity, air pressure, wind speed, CO2concentration, soil

water availability and by the trees themselves through stomatal

conductance [28], [29]. In freezing conditions, water within

the trees freezes, resulting in a signiﬁcant drop in permittivity.

Reversible freezing occurs at different times, temperatures and

rates in the various forest structures [30].

The mechanisms by which the permittivity of forest struc-

tures vary with time are complex and are currently not fully

understood. Our lack of knowledge of how trees respond to

their environment can be attributed to the laborious, indirect,

invasive and even destructive nature of measurement tech-

niques for quantifying spatio-temporal tree water content and

chemical concentrations. The effects of changing weather and

seasonal conditions on forest backscatter can currently only

be revealed by empirical studies. Existing P-band observation

data of boreal forests are limited to that of a few airborne

campaigns [31]–[35]. These observations lack the temporal di-

versity necessary for investigating the temporal characteristics

of P-band backscatter.

In this study, a boreal forest stand was observed using

a tower-based multi-polarimetric, tomographic P-band radar.

The aims of the study were to:

1) Collect densely-sampled time series of backscatter from

different height intervals within the forest canopy

2) Identify the most signiﬁcant temporal features in the

dataset

0 75 150 m

Radar tower

Observed forest

Corner reflectors

Fig. 1. RGB aerial photo of the experiment site acquired on 10 May 2018 by

the Swedish National Land Survey (Lantm¨

ateriet). The bottom corner reﬂector

was used for calibrating the tower radar and the top corner reﬂector was

temporarily installed for airborne SAR observations.

3) Gain insight into the relationship between backscatter,

meteorological variables and ecophysiological mecha-

nisms

The goal of the experiment is to gain a better understanding of

the electromagnetic scattering mechanisms taking place during

SAR observations and how SAR observations are affected by

changing weather and seasonal conditions.

First, the experiment is described in Section II and data anal-

ysis methods are detailed in Section III. Section IV contains

the ﬁrst result analysis of the dataset, then a discussion of the

results and their implications on forest parameter retrieval is

given in Section V.

II. EX PE RI ME NT D ES CR IP TION

A. Experiment site

The observed forest is a homogeneous, mature stand of

Norway spruce (Picea abies (L.) Karst). The forest stand is

located in the Remningstorp experimental forest in southern

Sweden and had an above-ground biomass density of 250

tons/ha in the fall of 2014 [4]. The terrain is ﬂat, is covered in

moss and has little understory. The canopy height varies from

25 to 27 m. The 50 m-high radar tower is located at the edge

of the forest stand (58◦27’ 5” N, 13◦37’ 35” E) as shown

in Fig. 1. Several trihedral corner reﬂectors are placed around

the site for calibration purposes.

B. Radar instrument

The radar instrument consists of a vector network analyzer

(VNA) connected to an antenna array (see Fig. 2). The VNA

has 20 ports, each of which is connected to one of the

20 antennas (10 transmit and 10 receive) in the array at

the top of the tower (see Fig. 3). The VNA operates by

transmitting, from a single VNA port, monochromatic pulses

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

VNA Microwave

switch

Diplexers

surge

protectors

Low

loss

cables

Antennas

Reference

cable

Temperature-controlled instrument hut Tower

Fig. 2. Block diagram of the radar instrument.

over a range of frequencies in discrete frequency steps. The

stepped-frequency sweep covers a bandwidth of 30 MHz

centered at 435 MHz. The signal is emitted in frequency

steps of 0.5 MHz. These signal parameters result in a range

resolution of 5 m and a maximum unambiguous range of

300 m. The signal is emitted as an electromagnetic wave

from a single antenna and scatters off the scene producing

a scattered wave. The scattered wave is sampled in space

by all 20 antennas in the array simultaneously. This parallel

measurement conﬁguration greatly reduces the measurement

time compared to systems employing mechanical switching

between antennas [36]. The antenna array was designed for

tomographic imaging of the forest scene below at P-band to

L-band (1270 MHz) for all linear polarization combinations

(HH, VV, HV and VH). A second array is used for C-

band (5410 MHz) measurements. Note that the design of the

antenna array implies different antenna patterns and phase

centers for the different polarization combinations. A speciﬁc

calibration procedure would be needed to exploit the polari-

metric phases and fully polarimetric data. Such non-trivial

calibration procedure would require a speciﬁc study, so that we

prefer to focus here on the intensities derived for the above-

mentioned combination of polarization states. Details of the

array designs are given in [36] and [37]. The close antenna

spacing and ﬁnite-bandwidth VNA measurements result in

mutual antenna coupling, which adds distortion to the received

signals. The mutual coupling component of the received signal

was suppressed, without affecting the spatial resolution, using

a novel procedure described in [38]. The microwave switch in

Fig. 2) is a mechanical switch used only for internal calibration

purposes using a reference cable.

C. Temporal stability of the radar instrument

Temporal variations in the total gain of the signal chain

must be insigniﬁcant compared to variations in the forest

backscatter if temporal variations in the forest backscatter

are to be studied. To minimize the inﬂuence of weather

conditions on the system response, all active electronics were

housed in a temperature-controlled hut, antennas with weather-

resistant radomes were selected and all outdoor connections

were sealed with vulcanizing tape. The tower was designed

and constructed for a maximum horizontal deﬂection at the

tower top of ±3 cm for wind speeds up to 17 m/s. To assess

the temporal stability of the VNA’s measurement response,

each transmitting port was connected to each receiving port

via the reference cable (see Fig. 2) at least once every hour.

An example of the insertion loss and phase shift measured

through the reference cable between two VNA ports is shown

in Fig. 4. An insigniﬁcant amount of magnitude and phase

P/L-band

array

C-band

array

Anemometer

Fig. 3. Photo of the top section of the radar tower. The two antenna arrays

were connected to the VNA on the ground via coaxial cables.

Fig. 4. An example of the insertion loss and phase shift by a reference cable

connected between a transmit-receive pair of VNA ports measured at 1-hour

intervals at 435 MHz. There was very little variation over the duration of

experiment, indicating that there was an insigniﬁcant amount of drift in the

VNA’s measurement response.

variation was observed, indicating that the VNA measurement

response was stable for the duration of the experiment.

To assess the temporal stability of the low-loss cables

and antennas on the tower, the mutual coupling components

between antennas and the response from a trihedral corner

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Fig. 5. Examples of the mutual coupling power and forest reﬂectivity

extracted from an HH range proﬁle. The bottom two plots show the corner

reﬂector magnitude and phase for the duration of the experiment. The

little variation in the mutual coupling power compared to forest reﬂectivity

variations indicates that there was an insigniﬁcant amount of systematic

temporal variation in the instrument’s gain. The corner reﬂector phase was

also very stable for the duration of the experiment.

reﬂector were analyzed, as was done in [4]. The corner

reﬂector was placed on an open ﬁeld and in direct line of sight

of the radar antennas. Fig. 5 shows the mutual coupling power

and forest reﬂectivity as measured from an HH-polarized

range proﬁle. The mutual coupling power shows signiﬁcantly

less temporal variation compared to the forest reﬂectivity,

indicating that the system’s gain response is stable enough for

studying forest backscatter variations. The magnitude response

from the trihedral corner reﬂector showed more variation

than that of the mutual coupling component, appearing to

coincide with freezing temperatures. These variations were

not observed in the mutual coupling power, nor in the VV

trihedral response, and thus it was concluded that ground-

trihedral reﬂections were the cause of the observed ﬂuctuations

in the HH trihedral response. Ground conditions had less

of an inﬂuence on VV observations because the Brewster

effect is signiﬁcant at the large incidence angle (76◦) at the

corner reﬂector’s position. The phase of the trihedral corner

reﬂector was stable for the duration of the experiment. These

observations indicate that temporal variations in the system

response were negligible compared to variations in the forest

reﬂectivity.

D. Radar measurement sequence

Tomographic image measurements were carried out in

bursts of four measurements. The four measurements in a burst

were separated by 5 s, and bursts were separated by 5 min (see

5 min

5 s

VV & HV

tomographic

measurement

HH & VH

tomographic

measurement

40 ms

Time

4 burst

Fig. 6. Timeline of the measurement sequence. Every 5 minutes, a burst

of four tomographic observations were made. Four tomograms for each

polarization were thus acquired every 5 minutes.

Fig. 6). Each of the four measurements in a burst covered a

tomographic measurement for each polarimetric combination

(HH, HV, HH and VH). VV and HV measurements were done

simultaneously. Likewise, HH and VH measurements were

done simultaneously. This is possible because the VNA was

capable of receiving signals from all 20 antennas in parallel.

The measurement time for a single tomographic image was 40

ms, which is short enough for the forest scene to be assumed

coherent during a tomographic measurement.

E. Meteorological observations

An on-site weather station measured air temperature, pres-

sure and relative humidity at heights of 2 m and 30 m

above ground. The 3D wind vector was measured by two

ultrasonic anemometers installed at the top of the tower (50

m above ground), 20-30 m above the tree tops. Precipitation

was measured by a heated rain gauge, which cannot distinguish

between rain and snow. Soil temperature and volumetric water

content were measured by time-domain reﬂectometry probes

within the upper 30 cm of the soil.

Two surveillance cameras were installed on the tower to

observe the state of snow on the canopy and ground. A photo

was taken from each camera at 5 minute intervals, coinciding

with the radar measurements. Snow depth and solar radiation

were not measured during this study.

The water vapor pressure deﬁcit (VPD), which is one of the

drivers of transpiration, was estimated from the observed air

temperature and relative humidity. The saturation pressure of

water vapor, which is necessary for estimating the VPD, was

estimated using the Goff-Gratch equation [39].

III. MET HO D

A. Tomographic imaging

Tomographic SAR is based on multiple observations of

the same scene. Radar observations acquired over a range

of azimuth and elevation angles are coherently combined

to construct a 3D distribution of the backscattered power

[40]. This is typically implemented by multiple passes of an

airborne or spaceborne radar with different incidence angles.

The antenna array in this study has a vertical aperture,

providing ﬁne resolution in the ground range-height plane

only. Tomographic images were formed in this plane. Azimuth

resolution is determined by the antenna gain patterns, resulting

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

in a variable resolution cell size throughout the image plane.

This variable resolution cell size introduces a systematic gain

which is dependent on the antenna gain patterns, antenna array

geometry and signal parameters. This systematic gain was

compensated for by normalizing the backscatter of each pixel

by the energy of the image impulse response function for the

respective pixel [38].

Tomographic images were constructed from the VNA mea-

surements using a backprojection algorithm [38]. To produce a

focused tomographic image, it is necessary that the systematic

phase differences between transmit-receive measurements con-

tributing to the backprojection sum are small. Such phase, and

also magnitude, imbalances were estimated and compensated

for using a trihedral corner reﬂector as an external reference

using a method detailed in [38]. This method is robust to

temporal variations in the reﬂector’s reﬂectivity (e.g. due to

time-variant ground-reﬂector contributions) as long as the

system’s response is stable, as was concluded in Section II-C.

Examples of tomographic images are shown in Fig. 7. Each

image exhibits speckle, which is a result of the coherent

addition of scattering contributions from several elements

within a resolution cell. Therefore, conclusions about the

forest backscatter should not be drawn by directly analyzing

the tomograms. The trihedral corner reﬂector lies on the

ground at a ground range of 207 m, where the HH and VV

tomograms show bright spots. The HV and VH images are

different because different bistatic antenna pairs were used in

the measurements. However, the trihedral corner reﬂector is

also slightly visible in the HV image, which indicates that

there may be some cross-talk leakage for HV measurements.

The reason for this observation is presently unclear, and the

HV images were therefore not analyzed in this study. Cross-

polarization data are therefore estimated in this study through

the VH channel, which would be equal to HV tomographic

images for a monostatic radar, assuming target reciprocity.

Different antennas were used for HV and VH measurements.

The main forest region of interest (ROI) lies within a ground

range of 20 to 70 m of the tower, covering similar incidence

angles as space-borne SARs (20◦to 55◦). This region, indi-

cated by the dotted rectangle in Fig. 7, is free from forest

edges and corner reﬂectors. The forest reﬂectivity distribution

for this region peaks at two heights for all polarizations: at

the upper canopy (10 to 30 m height) and at the ground level

(-10 to 10 m height). Reﬂections at the upper canopy level

are mainly due to volume scattering and apparent reﬂections

at the ground level are mainly due to the sum of direct

ground scattering and double-bounce scattering (ground-trunk

and trunk-ground) [15], [34]. These observations motivate a

study of the backscatter mainly within these regions. The forest

observed in this experiment is dense, resulting in a strong

canopy contribution and attenuating the ground contribution.

B. Regions of interest

Temporal variation of backscatter in the tomographic images

was studied using the incoherently integrated pixel intensities

in three ROIs as deﬁned in Table I. This was done to simplify

the analysis of temporal variations by reducing the dimen-

sionality of the data and, to obtain more accurate estimates of

Fig. 7. Tomographic image examples from 06:00 on 1 September 2018.

Height proﬁles of the image backscatter within the regions surrounded by

dotted lines are shown on the right of each tomogram. Backscatter peaks at

the ground level and upper canopy level. The solid white line shows a lidar-

derived canopy height estimate, the brown line shows the ground level and

the 50 m-high tower is illustrated to scale on the left in each tomogram.

the backscatter. The ROIs are shown in Fig. 8 in the image

plane. Estimation accuracy of the image backscatter must be

improved by averaging over ROIs because the tomographic

images exhibit speckle, whereby a large variance is associated

with the backscatter estimate for a single pixel [41]. The

backscatter within an ROI was incoherently integrated to

yield a lower-variance estimate (higher number of looks) of

the mean backscatter within the ROI. To further increase

the number of looks, the backscatter estimates from all four

consecutive measurements in a burst (Fig. 6) were averaged.

Under calm conditions, the four tomographic images are nearly

identical, and thus no gain in the number of looks is achieved.

In dynamic conditions, such as during wind or rain, these four

images will be different due to scattering changes between

image acquisitions, and the number of looks will be up to

four times the estimates in Table I. The result is a multilooked

backscatter time series for each polarization (HH, VV and VH)

in each ROI (full forest, canopy and ground) with a sampling

interval of 5 minutes. This procedure is illustrated in Fig. 9.

C. Time series

Six time series were analyzed: 3 polarizations (HH, VV

and VH) for each of the three ROIs (canopy, ground and

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

0 20 40 60 80 100

Ground range [m]

-10

Height [m]

Full forest ROI

Canopy ROI

Ground ROI

Lidar canopy height

Ground level

Antenna array

Fig. 8. Diagram (to scale) showing the three ROIs on the image plane in

relation to the tower and forest geometry. The tower and forest visualization

were produced by a terrestrial lidar scan in 2017 by the Swedish University of

Agricultural Sciences (SLU). The canopy height estimate is the 99th percentile

from an airborne lidar scan in 2014.

Jun 14, 12:00 Jun 15, 00:00 Jun 15, 12:00 2017

Time

HH full forest ROI images from a single burst

Smoothed series

HH full forest ROI time series

1 2 3 4

Fig. 9. Illustration of how time series were formed. The backscatter of four

tomographic images was incoherently averaged to produce a single sample in

a time series. The time series was further smoothed to reveal variations over

longer timescales.

full forest). The time series exhibit variations at timescales

of minutes (due to wind-induced movement) to years (due to

seasons). The longer the timescale of interest, the more the

time series must be temporally smoothed. For example, wind-

induced variations obscure diurnal variations, so wind-induced

variations must be smoothed to analyze diurnal variations (see

Fig. 9). Unless otherwise stated, samples were averaged over

4 hours (48 samples). The Savitzky-Golay ﬁlter was used

to suppress short-term variations while preserving transients

[42], which occur due to different mechanisms (e.g. rain and

freezing).

All times in this article are local solar times (UTC+54.5 min

for the site’s longitude). In local solar times, the sun is highest

in the sky at 12:00 every day. This is helpful when analyzing

biological responses to diurnal variations in solar radiation.

The backscatter time series were not calibrated in an ab-

solute manner using the trihedral corner reﬂector because its

response was affected by the conditions of the surrounding

TABLE I

DEFI NIT IO N OF ROI S IN TO MO GRA PH IC IM AGE S.

ROI

Ground

range

interval

Height

interval

Number

of looks

Canopy 20 to 70 m 10 to 30 m 12.8

Ground 20 to 70 m -10 to 10 m 9.3

Full forest 20 to 70 m -10 to 30 m 19.7

soil. Instead, a preliminary radiometric calibration was done

using an airborne SAR as a reference [43]. This calibration

was only based on HH observations, and thus the absolute

backscatter levels of different polarizations were not compared

in this study.

D. Statistical inference

The observations in this experiment are subjected to a low

number of looks. This is a consequence of the permissible

transmit bandwidth (30 MHz), the number of antennas in the

array (vertical array aperture) and the limited ﬁeld of view

from the tower. As a result, temporal variations in the spatially

averaged backscatter due to speckle may be larger than the

physical backscatter variations under study. In order to avoid

incorrect conclusions, the expected backscatter variations due

to speckle must be quantiﬁed and their effect in the temporal

domain must be understood. A simulation was used to estimate

the number of looks, derive conﬁdence intervals, study the

effect of speckle in the time domain and to develop a method

for detecting periods under which the observations show true

physical variations in backscatter.

Tomographic images of a uniformly-distributed cloud of

5000 point scatterers within the full forest ROI were simulated.

The simulation took into account the signal properties, antenna

gain patterns, antenna array geometry and image reconstruc-

tion method [38]. To estimate the number of looks, 1000

images were simulated, each with a different realization of

point scatterers. For each of the 1000 images, the backscatter

over the three ROIs deﬁned in Table I was estimated. These

backscatter estimates were then used to estimate the equivalent

number of looks [44]. Table I lists the estimated number of

looks for each ROI. Fully-developed speckle was assumed,

which is a reasonable assumption for forest canopies and the

large azimuth antenna beamwidth [41]. Even though the areas

of the canopy and ground ROIs are the same, the estimated

number of looks is lower for the ground region. This is

because the resolution in elevation is slightly poorer closer

to the ground compared to the canopy, which is a result of the

antenna array geometry.

The simulation results were also used to estimate 90%

conﬁdence intervals for backscatter variations due to speckle.

These are plotted in Fig 10. According to the simulation,

backscatter variations in the canopy ROI due to speckle lie

within a range of approximately 4 dB for 90% of observations.

Incoherently averaging over 4 bursts will increase the number

of looks by a factor of 4, decreasing the conﬁdence interval to

2 dB. This is assuming windy conditions, under which each

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

10 30 100 300 1000 3000

Number of looks

90% confidence interval [dB]

Theoretical confidence interval

Full forest ROI (simulated)

Canopy ROI (simulated)

Ground ROI (simulated)

Full forest ROI+burst

Canopy ROI+burst

Ground ROI+burst

Full forest ROI+burst+smoothing

Canopy ROI+burst+smoothing

Ground ROI+burst+smoothing

Worst case

Best case

without

smoothing Best case with

smoothing

Fig. 10. Theoretical conﬁdence intervals (90%) of backscatter observations

due to speckle. Averaging over bursts and smoothing time series over time

increases the numbers of looks, making the conﬁdence interval narrower.

burst will provide an independent sample. Smoothing over 4

hours will further reduce the conﬁdence interval to below

0.5 dB. These measures appear problematic for observing

backscatter variations on the order of 1 dB, such as diurnal

cycles [2], unless the conditions are windy and a high number

of looks is attained. However, these measures are based on the

Rayleigh fading model, which is normally applied to spatial

backscatter statistics (e.g. pixels in a SAR image). Extending

these statistics to the temporal domain, in which the same

forest site is observed over time, is not straightforward be-

cause geometric changes depend on environmental conditions.

Therefore, the simulation was used to study the statistical

characteristics of temporal variations.

The decrease in tree water content during hot days is

expected to cause a drop in the canopy backscatter around

noon. The radar cross section of each point scatterer in the

canopy ROI was modelled as a sinusoid with a peak-to-

peak magnitude of 1 dB, reaching a minimum at 12 pm.

A different random component (normally distributed with a

standard deviation of 0.2 dB) was added to the sinusoidal

component every 5 minutes to model unequal backscatter

variations in the forest. One realization of the radar cross

section for a single point scatterer is shown in Fig. 11. The

resulting smoothed backscatter time series from 50 different

point scatterer position realizations are shown in Fig 12 (top

left). Although there is a large variation when considering

multiple realizations of point scatterer positions (e.g. pixels

in a SAR image), the time series for any single geometric

realization shows the correct 1 dB backscatter variation.

To investigate the effect of wind on backscatter time series,

a random component (normally distributed with a standard

deviation of 0.5 m) was added to the point scatterer positions

for every tomogram simulated. This resulted in the smoothed

time series shown in Fig. 12 (top right). The decorrelation

caused by the ﬂuctuating geometry increases the number of

looks, decreasing the variance between the 50 different point

scatterer realizations. The time series for any forest realization

Fig. 11. Example of the radar cross section of a point scatterer (top) and

the standard deviation of point scatterer positions (bottom) used as simulation

inputs.

Fig. 12. Simulation results of backscatter time series from 50 different point

scatter scene realizations.

again shows the correct 1 dB backscatter variation.

Strong winds rarely persist for entire days. During the

summer, convective winds are common, in which the wind

speed is high during the day and low during the night. To

simulate diurnal wind speeds, a random component (normally

distributed) with a diurnally varying standard deviation was

added to the point scatterer positions. This standard deviation

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Fig. 13. Simulated relationship between the backscatter variance over a 4-

hour interval of unsmoothed time series and the 90% conﬁdence interval of

speckle variations for the canopy ROI.

is plotted in Fig. 11. The time series for a scene with this

diurnally varying geometry, but constant backscatter, is shown

in Fig. 12 (bottom left). During periods with large geometric

ﬂuctuations (high wind speeds), the variance between the

50 forest realizations is small. If the diurnal backscatter

component is added, the time series take the forms shown

in Fig. 12 (bottom right). The correct backscatter variations

are seen during the day when the geometry ﬂuctuates. During

the night, the backscatter will recover to a value with a large

amount of speckle variance.

The simulation results show that the correct temporal vari-

ations in backscatter are observed when:

Case 1 Geometric ﬂuctuations are small for several hours (top

left plot in Fig. 12).

Case 2 Geometric ﬂuctuations are large for several hours (top

right plot in Fig. 12).

When geometric ﬂuctuations vary signiﬁcantly during the day,

the time series might not show the true physical variation

in backscatter. The variance of unsmoothed observed time

series over a 4-hour period was used to determine whether the

time series show the correct physical variation in backscatter.

The simulated relationship between this variance and the 90%

conﬁdence interval of speckle variations is shown in Fig.

13. There is little change in the conﬁdence interval below a

backscatter variance of 0.1 dB, and thus 0.1 dB was adopted as

the upper threshold for detecting Case 1 in the observed time

series. During windy conditions, the 90% conﬁdence interval

of speckle variations is below 0.5 dB for a backscatter variance

of 0.38 dB and higher, which is sufﬁcient for observing

backscatter ﬂuctuations on the order of 1 dB. A variance of

0.38 dB was thus adopted as a lower threshold for detecting

Case 2. Segments of time series detected as Case 1 or Case 2

have a high likelihood of showing the true physical backscatter

variation.

IV. RES ULT S

The time series results in this section are presented in the

order of longest time scales and largest backscatter variations

to smaller backscatter variations at shorter time scales.

A. Long-term variations

The temporally-smoothed backscatter time series are shown

in Fig. 14 along with air temperature, precipitation and soil

moisture content. The backscatter from the canopy, ground

and full forest ROIs are plotted on the same axes for each

polarization. The largest temporal variations occurred during

the winters when the air temperature dropped below 0◦C. The

freeze-thaw backscatter variations were largest for the canopy

region, with more than 10 dB variations observed for all

polarizations. The full forest freeze-thaw variations for HH and

VV were similar to one another, differing mainly in amplitude.

HH variations during the winter (up to 8 dB) are larger than

those of VV (up to 5 dB). Winter variations in full forest VH

backscatter (up to 10 dB) were larger than those of the co-pol

channels because of the stronger contribution of the canopy

to the full forest backscatter. The HH and VV ground-level

backscatter time series were similar to those of the full-forest

backscatter since the ground backscatter variations dominated

the co-polarized full forest backscatter during the winters.

HH and VV full forest backscatter showed similar temporal

dynamics throughout the rest of the observation period as well.

Smaller variations over a range 3 to 5 dB at timescales of

weeks to months can be observed for HH and VV during

autumn 2017 and spring/summer 2018, which show some

correlation with soil moisture. VH does not show long-term

variations that are correlated with soil moisture. This is due to

the stronger double-bounce scattering at HH and VV, which

is affected by soil moisture. The ground-level backscatter

dynamics for HH and VV were similar to one another, showing

some correlation with soil moisture content during non-frozen

conditions. These ground level variations were larger for HH

compared to VV, likely as a result of stronger ground and

stem reﬂections at HH due to the Brewster effect. These ob-

servations conﬁrm that the co-polarized backscatter variations

are strongly inﬂuenced by double-bounce scattering, which

appears at the ground level. Soil moisture changes appear to be

the dominant cause of long-term variations in the ground-level

and full forest HH and VV backscatter during non-frozen con-

ditions. The ground-level VH backscatter differs signiﬁcantly

from the full forest VH backscatter. This is because ground-

level scattering, such as double-bounce scattering, is smaller

for VH and therefore has little inﬂuence on the full forest VH

backscatter. The ground-level VH backscatter showed some

freeze-thaw effects, but otherwise erratic variations. These

variations are ampliﬁed by the decibel scale since the ground-

level backscatter is relatively low for VH due to signiﬁcant

attenuation by the canopy and little double-bounce scattering.

During non-frozen conditions, the canopy backscatter was

stable on the long term for all polarizations, especially for HH

and VV. VH canopy backscatter does not show a signiﬁcant

improvement in stability compared to the full-forest backscat-

ter since the canopy contribution dominates the full forest

backscatter. The similarity in canopy-level backscatter dynam-

ics between polarizations indicates that the underlying scatter-

ing mechanism is independent of polarization. This mechanism

is volume scattering, in which the scattered electric ﬁeld is the

coherent sum of contributions of many, randomly orientated

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Fig. 14. Time series of the temporally-smoothed forest backscatter for the three ROIs and three polarization combinations. The air temperature, precipitation

and volumetric soil moisture content are shown in the bottom two plots.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Fig. 15. VH canopy backscatter and a 1st order Fourier series ﬁt showing

a seasonal component. Frozen conditions were manually removed from the

backscatter data.

scattering elements within a resolution cell. This makes the

canopy-level backscatter independent of soil moisture content.

The only signiﬁcant difference seen between polarizations is

that the VH canopy backscatter exhibits a seasonal component

with a peak-to-peak amplitude of approximately (2 dB) during

non-frozen conditions. This is shown in Fig 15. The cause

of this seasonal backscatter cycle, which is only visible for

VH, is not clear from the available data. Possible causes

could be seasonal moisture variations, tree ring formation

or reproduction cycles. The long-term periodicity makes it

unlikely that these cycles are a result of speckle variance.

Satellites in a dawn/dusk polar orbit will only observe

around 6:00 and 18:00. Fig. 16 shows the histograms of the

full forest backscatter for all observations, dawn only and

dusk only. Dawn observations included all observations be-

tween 5:30 and 6:30. Dusk observations included observations

from 17:30 to 18:30. For any given polarization, there is

no signiﬁcant difference in the backscatter distributions when

considering dawn or dusk only. This is because backscatter

variations over timescales longer than 1 day dominate the

full forest backscatter time series for all polarizations. In HH

and VV these variations appear to originate from changes

in the ground level backscatter. For VH, the source of long-

term variations is not clear and may partly be due to speckle

variance.

B. Effects of freeze-thaw cycles

Freeze-thaw cycles cause the largest variations in P-band

canopy backscatter. This is most clearly seen in the monthly

peak-to-peak backscatter range shown in Fig. 17. The canopy

backscatter range has a seasonal cycle, reaching up to 13 dB

during the winter. There is no signiﬁcant difference between

polarizations, except that the canopy backscatter range for VV

is lower than that of other polarizations during the early winter

of 2018-2019. This is likely due to the warmer temperatures

of the 2018-2019 winter.

The canopy backscatter is signiﬁcantly lower during freez-

ing temperatures, as is shown by the HH canopy backscatter

distributions in Fig. 18. This is because part of the free water

in the trees turns into ice, which has a signiﬁcantly lower

permittivity compared to liquid water [45], [46], resulting in

weaker reﬂections of electromagnetic waves. The distribution

HH (all)

-6 -4 -2 0 2

Backscatter [dB]

0.2

0.4

0.6

Probability

HH (dawn)

-6 -4 -2 0 2

Backscatter [dB]

0.2

0.4

0.6

Probability

HH (dusk)

-6 -4 -2 0 2

Backscatter [dB]

0.2

0.4

0.6

Probability

VV (all)

-8 -6 -4 -2

Backscatter [dB]

0.2

0.4

0.6

Probability

VV (dawn)

-8 -6 -4 -2

Backscatter [dB]

0.2

0.4

0.6

Probability

VV (dusk)

-8 -6 -4 -2

Backscatter [dB]

0.2

0.4

0.6

Probability

VH (all)

-16 -14 -12 -10 -8 -6

Backscatter [dB]

0.1

0.2

0.3

0.4

0.5

Probability

VH (dawn)

-16 -14 -12 -10 -8 -6

Backscatter [dB]

0.1

0.2

0.3

0.4

0.5

Probability

VH (dusk)

-16 -14 -12 -10 -8 -6

Backscatter [dB]

0.1

0.2

0.3

0.4

0.5

Probability

Fig. 16. Histograms showing the distributions of full forest backscatter when

considering all observations (left column), dawn observations only (middle

column) and dusk observations only (right column). Dawn was deﬁned as

from 5:30 to 6:30 and dusk was deﬁne as from 17:30 to 18:30.

Jun 17

Jul 17

Aug 17

Sep 17

Oct 17

Nov 17

Dec 17

Jan 18

Feb 18

Mar 18

Apr 18

May 18

Jun 18

Jul 18

Aug 18

Sep 18

Oct 18

Nov 18

Dec 18

Jan 19

Feb 19

Mar 19

Apr 19

May 19

Jun 19

Monthly backscatter range [dB]

Fig. 17. Monthly range of the canopy backscatter. The winter periods show

the largest range of canopy intensities due to freeze-thaw cycles and do not

vary signiﬁcantly between polarizations, except for VV in the winter of 2018-

2019.

of canopy backscatter during freezing temperatures is bi-

modal, showing that the vegetation does not freeze at the

exact onset of sub-zero air temperatures. This may be due

to a difference in air and tree temperatures or because free

water in vegetation may exist in a supercooled state before

ice nucleation occurs [30], [47], [48]. Canopy and full forest

backscatter were comparably stable during non-frozen condi-

tions, with a standard deviation of 0.7 to 1 dB.

A sample of the HH backscatter time series during the

winter is shown in Fig. 19 for closer inspection. The canopy

backscatter drops signiﬁcantly (up to 10 dB) when the air tem-

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Canopy

-15 -10 -5

Backscatter [dB]

0.1

0.2

0.3

0.4

0.5

0.6

Probability

Tair 0°C

Tair < 0°C

Full forest

-10 -5 0 5

Backscatter [dB]

0.1

0.2

0.3

0.4

0.5

0.6

Probability

Fig. 18. Histograms showing the HH backscatter distributions for tem-

peratures above and below 0◦C for the entire observation period. Other

polarizations show similar distribution differences.

Fig. 19. Canopy and ground-level tomographic backscatter for HH-

polarization during freeze-thaw cycles. Canopy backscatter drops signiﬁcantly

during freezing temperatures. Ground-level backscatter is stronger in freezing

temperatures because the canopy attenuation is lower. For sustained freezing

temperatures, the ground backscatter eventually decreases as it freezes from

the top down.

perature drops from positive to negative temperatures. Mois-

ture in the canopy freezes rapidly, making the canopy more

transparent to a P-band radar. This is true for all polarizations.

The reduced canopy attenuation during freezing temperatures

results in an increase in ground-level backscatter at the onset

of negative temperatures. Canopy and ground-level backscatter

for HH and VV counteract one another during freeze-thaw

cycles. As a result, the full forest backscatter, which is the sum

of the canopy and ground-level reﬂections, shows little freeze-

thaw variation for much of the winter (see Fig. 14). However,

during sustained negative air temperatures, the ground and

lower tree trunks gradually freeze, reducing the double-bounce

scattering and lowering the ground-level backscatter (Feb-

March 2018 in Fig. 19). This leads to large drops in full-

forest backscatter during the winter. Freeze-thaw cycles for

HH and VV have a signiﬁcant impact on tomographic P-band

SAR, but the full-forest backscatter is only affected during

sustained negative air temperatures. For VH, canopy scattering

dominates the full-forest backscatter, causing large variations

in both canopy and full-forest backscatter during freeze-thaw

cycles.

A strong positive correlation is observed between canopy

Fig. 20. Scatterplots of the canopy backscatter vs. air temperature. For

temperatures below 0◦C, a strong correlation between air temperature and

canopy backscatter exists due to varying ice fractions in the trees

backscatter and air temperature during freezing conditions in

the scatterplots in Fig. 20. The dielectric properties of ice have

a negligible temperature dependence below 0◦C [49], [50],

and do not cause the observed backscatter variations. Instead,

these observations are due to a varying fraction of free water

in the trees which turns into ice [51]. The ice fraction of wood,

and thus its permittivity, is strongly temperature dependent for

sub-zero temperatures [52].

The freeze-thaw dynamics of the ground-level backscatter

are more complicated because canopy freezing, stem freezing

and ground freezing do not necessarily occur at the same

time. Spatio-temporal ice formation mechanisms in trees are

not well understood and the role of ground freezing in the

observed freeze-thaw dynamics is not clear from this study.

Our results agree with previous studies that the canopy freezes

ﬁrst [30], [53], after which the lower trunk gradually freezes

over the timescale of hours. Ground backscatter exhibits

polarization-dependent hysteresis during freeze-thaw cycles

[?]. However, the freezing dynamics in the trunks and soil

cannot be separated in tomographic imaging, complicating

the modelling and interpretation of ground-level freeze-thaw

dynamics.

C. Effects of snow cover and melting snow

The upper canopy backscatter does not appear to be affected

by snow. Surveillance camera footage showed that snow

collecting on branches caused the branches to sag. This change

in geometry had no observable effect on the upper-canopy

backscatter since the branches were mostly transparent to P-

band radar during sub-zero air temperatures. The upper canopy

backscatter changes were instead more clearly correlated with

temperature variations, which through the canopy attenuation

affects the ground-level backscatter. Snow that collected on the

ground (10 cm) also had no distinguishable effect on ground-

level backscatter.

The effect of melting snow is difﬁcult to examine since

it is usually accompanied by an air temperature change

from negative to positive degrees Celsius, which dominates

backscatter variations. Rain is often the cause of snow melting

on the ground, which causes a clear increase in ground-level

backscatter in channels where the ground plays a signiﬁcant

role (HH and VV). HH and VV backscatter were closely

correlated with soil moisture content when the snow was

melting. Snow on the ground melting in the absence of rain did

not show any signiﬁcant change in ground-level backscatter.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

These results suggest that snow has no signiﬁcant effect on

P-band radar observations of the studied forest site. However,

this site did not experience long-term, heavy snowfall during

the observation period, leaving these results inconclusive.

D. Diurnal variations

Backscatter variations on diurnal timescales were observed

during the summers. Fig. 21 shows an example of diurnal

backscatter cycles for HH during the hot summer of 2018,

along with the VPD. The canopy backscatter was consistently

lower during the day and higher during the night. The period

shown in Fig. 21 was mostly detected as Case 1, implying

that the unsmoothed canopy backscatter time series had a

variance that was small enough such that speckle variations

over time are negligible (see Section III-D). Therefore, there

is a high likelihood that the canopy backscatter time series

show true physical variations for this period. Similar diurnal

backscatter variations were observed for all polarizations (see

Fig. 22), during all three summers, and in a tropical forest

by the TropiScat experiment [2], further suggesting that the

observed variations are true physical variations.

The VPD reaches high values around noon, which is

an indication of a high rate of transpiration. During these

high rates of transpiration, a decrease in canopy backscatter

is observed, increasing again in the evening. This canopy

backscatter decrease is likely due to a decrease in tree water

content in the upper canopy. The diurnal patterns for HH and

VV were very similar, suggesting similar diurnal scattering

mechanisms. The HH and VV ground backscatter showed a

sharp increase around noon, which may be connected with

stomatal closure. Stomatal closure occurs when the VPD

is very high and trees risk undergoing permanent damage

whereby the water columns in xylem vessels break apart due to

low pressures, becoming air-ﬁlled (cavitation) [55]. To regulate

this pressure, trees may close their stomata in the middle of

the day, limiting transpiration and halting the decrease in tree

water content [56]. Midday stomatal closure may also be the

cause of the bump around noon in the canopy backscatter seen

in Fig. 22.

Diurnal cycles, as are shown in Figs. 21 and 22, were

most clearly visible during periods with favorable atmospheric

conditions for transpiration. Such conditions include warm

air temperatures, low relative humidity, unobstructed sunlight

and moderate-to-high wind speeds [28], [29]. Diurnal cycles

were not observed during the autumns and winters. Diurnal

backscatter variations with similar peak-to-peak amplitudes (1

dB) and the same diurnal phases were observed in a tropical

forest for HH, VV and HV, but only during the dry season [2].

In hot and dry conditions, the rate of transpiration exceeds the

rate of water uptake by the roots in the morning, resulting in a

depletion of the tree’s internal water reserves [26], [27], [57].

The water reserves are replenished again during the evening

and night through water uptake by the roots when the rate

of transpiration decreases, i.e. the rate of root water uptake

exceeds the rate of transpiration. The tree water reserves

are believed to serve as a buffer between root water uptake

and water released to the atmosphere through transpiration to

Fig. 21. Diurnal cycles in HH backscatter during the summer of 2018.

A high VPD indicates high rates of transpiration, making the canopy lose

moisture. This reduces the backscattered power from the canopy and reduces

canopy attenuation, increasing double-bounce scattering seen at the ground

level. The tick marks corresponds to midnight each day. The thick lines in

the top plot mark time series segments detected as Case 1 (see Section III-D)

where speckle variance is low and the observed temporal variations are likely

due to a change in the scattering intensity of all scatterers in the region of

interest.

avoid permanent damage by cavitation [26]. Water reserves are

believed to reside in the bark and sapwood. It is the decrease

of water within these tissues which may be the cause of the

observed decrease in canopy backscatter during hot days. The

reﬁlling of water reserves during the evenings and nights may

then be the cause of the higher backscatter during these times

compared to during the day. These observations indicate that

P-band radar measurements of forests are sensitive to the

diurnal transpiration cycles in boreal forests and that, under

certain environmental conditions, biophysiological variables

such as transpiration rate and tree water content may be sensed

using P-band radar.

E. Effects of rain

During rainfall, much of the rain is intercepted by the forest

canopy in this dense forest. This increase in water on the

canopy can be expected to have an effect on the observed

backscatter. Studies have shown that rainfall can also lead to

an increase in the permittivity of trees, while some rainfall

events have no effect [58], [59]. The effect of rain during

the summer of 2018 is shown in Fig. 23. During the three

heavy rainfall periods (A, B and C in Fig. 23), the canopy

backscatter increased notably for HH and VV (1 to 2 dB).

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Fig. 22. Diurnal cycles for the summer period of 17 May 2018 to 9 August

2018. All polarizations show a decrease in canopy backscatter during the day

when the transpiration rate is highest. Only HH and VV show a clear ground

backscatter increase around noon. The daily mean was subtracted from each

daily segment before producing the statistics shown.

This suggests that the intercepted rainfall increases canopy

backscatter due to an increase in water on the canopy. One

can also expect an increase in canopy attenuation, leading

to a decrease in ground-level backscatter during heavy rain.

Such a drop in ground-level backscatter was only clearly

seen for VV in Fig. 23, whereas the ground-level backscatter

for HH seemed to increase by a small amount during rain.

The rainfall events were accompanied by large increases in

soil moisture at the ground surface. Such an increase in

soil moisture should affect both HH and VV ground-level

backscatter with approximately the same magnitudes since the

soil Fresnel reﬂection coefﬁcients for both horizontally- and

vertically-polarized waves at oblique incidence increase by

approximately the same magnitudes with an increase in soil

moisture [41]. The difference between the behavior of HH

and VV ground-level backscatter during heavy rainfall may

be due to different canopy attenuation mechanisms, whereby

heavy rainfall increases the canopy attenuation for VV more

than for HH, but speckle may also have a dominant effect

A B C D E

Fig. 23. Canopy, ground-level and full forest backscatter time series along

with the rainfall. The dashed blue lines and labels A to E mark the

times at which rainfall occurs. The time series were smoothed to remove

diurnal variations. Heavy rain has a clear effect on canopy and ground-level

backscatter whereas light rain does not have any signiﬁcant effect.

during rainfall.

The light rainfall events (D and E in Fig. 23) did not have

any signiﬁcant effect on the observed backscatter. Since heavy

rainfall such as events A, B and C in Fig. 23 were rare in this

region, rain rarely had any noticeable effect on tomographic

image backscatter at P-band. Events A, B and C in Fig. 23

should therefore be considered as extreme cases.

F. Effects of wind

Wind causes geometric changes in the observed scene

as trees sway in the wind. These geometric changes cause

ﬂuctuations in the observed backscatter. Fig. 24 shows the

observed canopy backscatter with and without temporal av-

eraging (smoothing) as well as the average wind speed for a

period during the 2017 summer. During periods with strong

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

winds (>5 m/s), large backscatter ﬂuctuations over timescales

of hours can be seen. This is temporal speckle, or Rayleigh

fading, which occurs when the scattered ﬁelds from scattering

elements in the forest interfere constructively and destruc-

tively over time. This is true for all polarizations. During

summer nights and mornings, there was little wind, making

the backscatter very stable over timescales of hours.

The temporally-averaged backscatter time series in Fig. 24

also show that strong winds coincide with a 1 to 2 dB drop in

average canopy backscatter. This is most common in segments

detected as Case 2, meaning that the wind causes a large

amount of temporal speckle and that the smoothed time series

show the true physical backscatter variations larger 0.5 dB

(see Section III-D). Similar to diurnal cycles observed during

windless conditions, these drops in backscatter usually occur

during the day, with the backscatter increasing again during

the night, even during Case 2 conditions (e.g. 30 June 2017).

The observed drops in the smoothed canopy backscatter time

series are, therefore, not due to speckle. A similar drop in full

forest backscatter was previously observed in non-tomographic

measurements [4], which was attributed to decreased double-

bounce scattering as the trees bend in the wind. The motivation

for this interpretation was that the drop in backscatter was not

seen for VH polarization, which is less sensitive to double-

bounce scattering. However, the tomographic data shows that

a drop in the canopy backscatter during high wind speeds, as

in Fig. 24, occurs for all polarizations. Since the relationship

between canopy backscatter and wind speed is not one-to-one,

they are best compared with their rates of change with time,

as is shown in Fig. 25. An increase in wind speed appears to

cause a decrease in the canopy backscatter of HH, VV and VH,

and an increase in the ground backscatter of VH. The ground

backscatter of HH and VV was insensitive to wind speed. This

increase in VH ground backscatter is likely due to increased

double-bounce scattering between the bending stems and the

ground [60], counteracting the decrease canopy backscatter

during windy conditions. This explains why the effect was not

seen in [4] for cross-polarized full forest backscatter. These

results show that a change in double-bounce scattering is

not the sole underlying mechanism behind the decrease in

backscatter during strong winds. A more likely cause of the

observed drop in canopy backscatter is a decrease in water

content in the canopy during windy summer days. When trees

transpire, the air at the leaf/needle surface becomes saturated

with water vapor from the stomata, decreasing the VPD at the

leaf surface and reducing the rate of transpiration [56], [61].

Wind replaces this saturated air with dry air, increasing the

VPD at the stomata and increasing the rate of transpiration

[62], [63]. The high rate of transpiration during strong winds

decreases the tree water content, causing a decrease in canopy

backscatter during the day.

These results suggest that wind causes a decrease in the

mean canopy backscatter with the same mechanism causing

diurnal cycles during hot periods: high rates of transpiration

causes a depletion of the tree’s water reserves during the

day, decreasing the backscatter. This theory is supported by

the observation that strong winds during cold conditions did

not cause a decrease in mean canopy backscatter. Strong

winds during periods with little solar radiation will cool the

stomata, decreasing the rate of transpiration [62]. The rate

of transpiration will also be low if the relative humidity is

high, such as during rainy periods. During cold or humid

conditions, depletion of the tree’s internal water reserves is

not necessary to sustain the low rate of transpiration driven

by the atmospheric conditions.

V. DI SC US SI ON

The largest backscatter variations observed were caused

by freezing temperatures. The large drops in canopy (as

well as full forest) backscatter during frozen conditions may

result in large errors in forest parameter estimates and forest

disturbance detections using P-band SAR data. The multi-

decibel backscatter variations observed during the winter will

result in boreal forest biomass estimation errors well above the

speciﬁed 20% for BIOMASS [64]. Correcting for this drop is

not simple since the permittivity of wood is strongly dependent

on temperature during frozen conditions. The best approach

would be to discard SAR observations acquired during frozen

conditions, but this requires frozen conditions to be detected.

Such a detection is complicated by the observation that the

drop in backscatter does not occur until a few degrees below

0◦C, and thawing may not occur until several degrees above

0◦C. Therefore, air temperature cannot be used as a reliable

proxy. Another complication is that while the canopy may be

frozen, the effect might not be seen in the full forest HH and

VV backscatter as observed by a single SAR overpass. How-

ever, this was not the case for cross-polarized observations.

Therefore, the best possibility for detecting frozen conditions

from backscatter data would be to detect changes (>2 dB) in

cross-polarized backscatter between SAR overpasses. Large

drops in cross-polarized backscatter, which do not appear in

co-polarized backscatter, are likely due to frozen trees during

the time of overpass and not deforestation. However, during

sustained frozen conditions, the co-polarized channels will

also show a drop in backscatter (see Fig. 14). Therefore,

this approach is questionable for other forest densities and

climates. The possibility of using phase shifts for detecting

freezing conditions should be investigated in the future.

HH and VV full forest backscatter were inﬂuenced by

large variations in ground-level backscatter (3 to 5 dB) due

to soil moisture changes. HH appeared most sensitive to soil

moisture variations. The corresponding variations in the full

forest backscatter were smaller since the canopy contributed

signiﬁcantly to the full forest backscatter in this dense forest.

In forests of lower density, with less canopy backscatter and

attenuation, soil moisture variations can be expected to play

a larger role in full forest HH and VV backscatter. Canopy

backscatter was very stable over long timescales if frozen pe-

riods are excluded (see Fig. 14). Therefore, canopy backscatter

will be accurately rendered by repeat-pass SAR tomography

and ground-notching interferometry, assuming high temporal

coherence. These results support the current approaches for

biomass estimation using P-band SAR.

A 0.5 dB difference in canopy backscatter can be expected

between 6 am and 6 pm overpasses (see Fig. 22). The long

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Fig. 24. Time series of the canopy backscatter and wind speed during the 2017 summer. Strong winds cause both short-term ﬂuctuations in backscatter and

a drop in mean backscatter. Regions detected as Case 1 and Case 2 (see Section III-D) are shown by magenta and blue curves respectively.

Fig. 25. Time derivative of the mean backscatter vs. the derivative of

wind speed. The correlation coefﬁcient r is given in each plot. The canopy

backscatter decreases when wind speed increases. The opposite behavior is

seen for VH ground-level backscatter.

revisit times of polar-orbiting, sun-synchronous SARs makes

them unable to capture the full diurnal dynamics of forest

backscatter. Diurnal variations in canopy backscatter were

connected to the current understanding of tree water relations.

The canopy backscatter diurnal cycles resemble the temporal

signatures of stem radii as measured by dendrometers. These

cycles can be described as a constant value that dips during

the day. However, radar backscatter is more closely related

to permittivity than stem radius. Diurnal variations in stem

permittivity are more sinusoidal than those measured by den-

drometers [46], [58]. This discrepancy may partially be ex-

plained by the nonlinear relationship between permittivity and

backscatter [20]. The relationship between canopy backscatter

and the permittivity measured at a single point in a stem may

also be more complicated than what is currently accepted. To

the authors’ knowledge, no experimental work has been carried

out in which both backscatter and permittivity were measured

over diurnal timescales. Time series of the in situ permittivity

of upper canopy structures have also not been acquired yet.

Short-term ﬂuctuations in backscatter due to wind are not

expected to signiﬁcantly affect P-band SAR observations.

Spatial multilooking in SAR images can be used to average

and reduce these ﬂuctuations. However, repeat-pass techniques

such as interferometry and tomography will be affected by

these ﬂuctuations in the form of temporal decorrelation. It

was also observed that strong winds may cause a canopy

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

backscatter drop of up to 2 dB during warm periods. These

drops in mean backscatter coincided with a large amount of

speckle variation. A multi-channel speckle ﬁlter, based on

samples from multiple times and polarizations was also tested

[65]. The smoothed time series were not signiﬁcantly affected

by this ﬁlter due to the dense multitemporal averaging capacity

of the data. Therefore, the ﬁlter was not included in the results

presented in this paper. The observed drops in backscatter may

cause a difference in the backscatter observed during 6 am

overpasses compared to 6 pm overpasses and may result in

minor errors in forest parameter estimates. However, strong

winds during hot conditions are rare in boreal forests.

VI. CO NC LU SI ON S

In this study, a mature stand of Norway spruce was moni-

tored by a multi-polarimetric, tomographic tower-based radar

over a two-year period. Time series of the forest backscatter,

ground-level backscatter and upper-canopy backscatter were

extracted from tomographic images and analyzed in relation

to meteorological variables. The largest temporal variations in

backscatter were due to freezing temperatures in the winter.

Such variations will lead to large forest parameter estimates

if not compensated for or excluded. The canopy backscatter,

which is the most sensitive to forest biomass, showed the

largest changes in backscatter (10 dB drops) during freezing

temperatures. Ground backscatter only dropped during sus-

tained freezing conditions. During non-frozen conditions, the

canopy backscatter was observed to be more stable than the

ground and backscatter for all polarizations due to decreased

sensitivity to soil moisture changes. Cross-polarized full forest

backscatter was dominated by the canopy contribution, show-

ing similar temporal variations during non-frozen conditions.

Observations during dawn or dusk times did not show a

decrease in temporal backscatter variations. Diurnal variations

(1 dB) were observed during hot periods and even larger

variations (up to 2 dB) were observed during strong winds

in the summer. These phenomena appeared to be connected to

high transpiration rates that result in a decrease in tree water

content during the day.

Regarding the diurnal backscatter variations observed, in-

formation about the water transport in forests is important

for ecophysiology, soil water dynamics and the management

of watersheds. Tower-based radars could be used as a new

research tool for characterizing tree water content signatures

over spatial scales of forest stands to ecosystems. Changes in

tree water content can be used to constrain canopy conduc-

tance, which is a key variable in soil-vegetation-atmosphere

transport models, ecosystem productivity models and global

climate models [66]. Although only a single tree species was

observed in this study, trees of different species have very

similar diurnal characteristics of water transport [67], and

can be expected to show similar backscatter signatures. The

magnitude of backscatter variations due to tree water content

variations is still uncertain and thus, our interpretation of the

observed diurnal cycles is speculative.

It is concluded that several questions remain to be ad-

dressed. The origin of canopy-level backscatter is not clear for

the different polarizations. A decrease in canopy backscatter

did not always appear to coincide with an increase in ground

backscatter (e.g. through a reduction of canopy attenuation).

Separation of backscatter contributions from different regions

in the forest is difﬁcult even with tomographic radar. Changes

in ground-level backscatter, which has a signiﬁcant inﬂuence

on HH and VV full forest backscatter, can be caused by canopy

attenuation, stem water content, or soil moisture content

variations. These effects could not be separated in this study,

complicating the interpretation of the effects of freeze-thaw

cycles, strong winds and rain. It is also concluded that future

work should include the analysis of backscatter data along

with dendrometer, sap ﬂow, and soil moisture proﬁle data.

Measurements of the in situ permittivity will also be made

since there is a possibility that the stem water content and

permittivity can be affected by chemical changes [58], [68].

Measurements of the complex permittivity will also clarify

the relationship between backscatter and attenuation in forest

canopies. Finally, a similar analysis for the phase evolution

over time and for observations at other frequency bands (e.g.

L-band and C-band) will be done, providing deeper insight

into the electromagnetic scattering mechanisms and temporal

backscatter variations throughout the forest canopy.

ACK NOW LE DG ME NT

The authors would like to thank the Swedish Defence Re-

search Agency (FOI) for providing and installing the trihedral

corner reﬂectors at the experimental site, the Swedish Univer-

sity of Agricultural Sciences (SLU) for providing the terrestrial

lidar point cloud of the forest site and the European Space

Agency (ESA) for funding the BorealScat experiment. We

also thank the two anonymous reviewers for their constructive

input.

REF ER EN CE S

[1] C. Albinet, P. Borderies, N. Floury, and E. Pottier, “Measure of temporal

variation of P-band radar cross section and temporal coherence of a

temperate tree,” IEEE Transactions on Geoscience and Remote Sensing,

vol. 54, no. 11, pp. 6255–6264, 2016.

[2] A. Hamadi, C. Albinet, P. Borderies, T. Koleck, L. Villard, D. Ho

Tong Minh, and T. Le Toan, “Temporal survey of polarimetric P-band

scattering of tropical forests,” IEEE Transactions on Geoscience and

Remote Sensing, vol. 52, no. 8, 2014.

[3] J. T. Pulliainen, L. Kurvonen, and M. T. Hallikainen, “Multitemporal

behavior of L- and C-band SAR observations of boreal forests,” IEEE

Transactions on Geoscience and Remote Sensing, vol. 37, no. 2 II, pp.

927–937, 1999.

[4] A. R. Monteith and L. M. H. Ulander, “Temporal survey of P- and L-

band polarimetric backscatter in boreal forest,” IEEE Journal of Selected

Topics in Applied Earth Observations and Remote Sensing, vol. 11,

no. 10, pp. 3564–3577, 2018.

[5] ESA (2012), Report for Mission Selection: BIOMASS. ESA SP-

1324/1 (3 volume series), European Space Agency, Noordwijk, The

Netherlands.

[6] S. Quegan, T. Le Toan, J. Chave, J. Dall, J.-F. Exbrayat, D. H. T. Minh,

M. Lomas, M. M. d’Alessandro, P. Paillou, K. Papathanassiou, F. Rocca,

S. Saatchi, K. Scipal, H. Shugart, T. Smallman, M. J. Soja, S. Tebaldini,

L. M. H. Ulander, L. Villard, and M. Williams, “The European Space

Agency BIOMASS mission: Measuring forest above-ground biomass

from space,” Remote Sensing of Environment, vol. 227, pp. 44–60, 2019.

[7] T. Le Toan, A. Beaudoin, J. Riom, and D. Guyon, “Relating forest

biomass to SAR data,” IEEE Transactions on Geoscience and Remote

Sensing, vol. 30, no. 2, pp. 403–411, 1992.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

[8] M. C. Dobson, F. T. Ulaby, T. Le Toan, A. Beaudoin, E. S. Kasischke,

and N. Christensen, “Dependence of radar backscatter on coniferous

forest biomass,” IEEE Transactions on Geoscience and Remote Sensing,

vol. 30, no. 2, pp. 412–415, 1992.

[9] H. Israelsson, J. Askne, and R. Sylvander, “Potential of SAR for

forest bole volume estimation,” International Journal of Remote Sensing,

vol. 15, no. 14, pp. 2809–2826, 1994.

[10] S. H. Alemohammad, T. Jagdhuber, M. Moghaddam, and D. Entekhabi,

“Soil and vegetation scattering contributions in L-band and P-band

polarimetric SAR observations,” IEEE Transactions on Geoscience and

Remote Sensing, vol. 57, no. 11, pp. 8417–8429, 2019.

[11] F. Banda, D. Giudici, T. Le Toan, M. Mariotti d’Alessandro, K. Pap-

athanassiou, S. Quegan, G. Riembauer, K. Scipal, M. J. Soja, S. Tebal-

dini, L. M. H. Ulander, and L. Villard, “The BIOMASS Level 2

prototype processor: Design and experimental results of above-ground

biomass estimation,” Remote Sensing, vol. 12, no. 6, 2020.

[12] J. M. B. Carreiras, S. Quegan, T. Le Toan, D. H. T. Minh, S. Saatchi,

N. Carvalhais, M. Reichstein, and K. Scipal, “Coverage of high biomass

forests by the ESA BIOMASS mission under defense restrictions,”

Remote Sensing of Environment, vol. 196, pp. 154 – 162, 2017.

[13] M. J. Soja, G. Sandberg, and L. M. H. Ulander, “Regression-based

retrieval of boreal forest biomass in sloping terrain using P-band

SAR backscatter intensity data,” IEEE Transactions on Geoscience and

Remote Sensing, vol. 51, no. 5, pp. 2646–2665, 2013.

[14] D. H. T. Minh, T. Le Toan, F. Rocca, S. Tebaldini, L. Villard, M. R´

ejou-

M´

echain, O. L. Phillips, T. R. Feldpausch, P. Dubois-Fernandez, K. Sci-

pal, and J. Chave, “SAR tomography for the retrieval of forest biomass

and height: Cross-validation at two tropical forest sites in french guiana,”

Remote Sensing of Environment, vol. 175, pp. 138–147, 2016.

[15] S. Tebaldini and F. Rocca, “Multibaseline polarimetric SAR tomography

of a boreal forest at P- and L-bands,” IEEE Transactions on Geoscience

and Remote Sensing, vol. 50, no. 1, pp. 232–246, 2012.

[16] M. J. Soja, M. M. d’Alessandro, S. Quegan, S. Tebaldini, and L. M. H.

Ulander, “Model-based estimation of tropical forest biomass from notch-

ﬁltered P-band Sar backscatter,” IGARSS 2018 - 2018 IEEE International

Geoscience and Remote Sensing Symposium, pp. 8617–8620, July 2018.

[17] M. J. Soja, S. Quegan, M. M. d’Alessandro, F. Banda, K. Scipal,

S. Tebaldini, and L. M. Ulander, “Mapping above-ground biomass in

tropical forests with ground-cancelled P-band SAR and limited reference

data,” Remote Sensing of Environment, Article in press, 2020.

[18] F. T. Ulaby, “Radar response to vegetation,” IEEE Transactions on

Antennas and Propagation, vol. 23, no. 1, pp. 36–45, 1975.

[19] E. Puttonen, C. Briese, G. Mandlburger, M. Wieser, M. Pfennigbauer,

A. Zlinszky, and N. Pfeifer, “Quantiﬁcation of overnight movement of

birch (betula pendula) branches and foliage with short interval terrestrial

laser scanning,” Frontiers in Plant Science, vol. 7, no. FEB2016, 2016.

[20] M. Moghaddam, S. Durden, and H. Zebker, “Radar measurement of

forested areas during OTTER,” Remote Sensing of Environment, vol. 47,

no. 2, pp. 154–166, 1994.

[21] D. Solimini, Understanding Earth Observation: The Electromagnetic

Foundation of Remote Sensing. Springer International Publishing

Switzerland, 2016.

[22] M. T. Hallikainen, F. T. Ulaby, M. C. Dobson, M. A. El-Rayes, and L.-K.

Wu, “Microwave dielectric behavior of wet soil-part I: Empirical models

and experimental observations,” IEEE Transactions on Geoscience and

Remote Sensing, vol. GE-23, no. 1, pp. 25–34, 1985.

[23] N. R. Peplinski, F. T. Ulaby, and M. C. Dobson, “Dielectric Properties

of Soils in the 0.3-1.3-GHz Range,” IEEE Transactions on Geoscience

and Remote Sensing, vol. 33, no. 3, pp. 803–807, 1995.

[24] M. Moghaddam, S. Saatchi, and R. H. Cuenca, “Estimating subcanopy

soil moisture with radar,” Journal of Geophysical Research Atmospheres,

vol. 105, no. D11, pp. 14 899–14 911, 2000.

[25] H. H. Dixon and J. Joly, “On the ascent of sap,” Philosohpical Trans-

actions of the Royal Society B, vol. 186, pp. 563–576, 1895.

[26] R. Zweifel, H. Item, and R. H¨

asler, “Link between diurnal stem radius

changes and tree water relations,” Tree Physiology, vol. 21, no. 12-13,

pp. 869–877, 2001.

[27] S. Pfautsch, T. H¨

oltt¨

a, and M. Mencuccini, “Hydraulic functioning of

tree stems - fusing ray anatomy, radial transfer and capacitance,” Tree

Physiology, vol. 35, no. 7, pp. 706–722, 2015.

[28] P. G. Jarvis, “The interpretation of the variations in leaf water potential

and stomatal conductance found in canopies in the ﬁeld,” Philosophical

Transactions of the Royal Society of London. B, Biological Sciences,

vol. 273, no. 927, pp. 593–610, 1976.

[29] T. Kumagai, Forest Hydrology and Biogeochemistry: Synthesis of Past

Research and Future Directions. Springer, Dordrecht, 2011, ch.

Transpiration in Forest Ecosystems.

[30] G. Charrier, M. Nolf, G. Leitinger, K. Charra-Vaskou, A. Losso,

U. Tappeiner, T. Am´

eglio, and S. Mayr, “Monitoring of freezing dynam-

ics in trees: A simple phase shift causes complexity,” Plant Physiology,

vol. 173, no. 4, pp. 2196–2207, 2017.

[31] I. Hajnsek et al., “BioSAR 2007: Technical assistance for the devel-

opment of airborne SAR and geophysical measurements during the

BioSAR 2007 experiment: Final report,” European Space Agency (ESA),

Paris, France, Tech. Rep. Contract No. 20755/07/NL/CB, 2008.

[32] ——, “BioSAR 2008: Technical assistance for the development of

airborne SAR and geophysical measurements during the BioSAR 2008

experiment: Final report,” European Space Agency (ESA), Paris, France,

Tech. Rep. Contract No. 22052/08/NL/CT, 2010.

[33] L. M. H. Ulander et al., “BioSAR 2010: Technical assistance for the

development of airborne SAR and geophysical measurements during the

BioSAR 2010 experiment: Final report,” European Space Agency (ESA),

Paris, France, Tech. Rep. Contract No. 4000102285/10/NL/JA/ef, 2011.

[34] M. Moghaddam and S. Saatchi, “Analysis of scattering mechanisms in

SAR imagery over boreal forest: Results from BOREAS ’93,” IEEE

Transactions on Geoscience and Remote Sensing, vol. 33, no. 5, pp.

1290–1296, 1995.

[35] S. Hensley, B. Chapman, M. Lavalle, B. Hawkins, B. Riel, T. Michel,

R. Muellerschoen, Y. Lou, and M. Simard, “UAVSAR L-band and P-

band tomographic experiments in boreal forests,” International Geo-

science and Remote Sensing Symposium (IGARSS), vol. 2018-July, pp.

8679–8682, 2018.

[36] L. M. H. Ulander, A. R. Monteith, M. J. Soja, and L. E. B. Eriksson,

“Multiport vector network analyzer radar for tomographic forest scat-

tering measurements,” IEEE Geoscience and Remote Sensing Letters,

vol. 15, no. 12, pp. 1897–1901, 2018.

[37] H. T. M. Dinh, S. Tebaldini, F. Rocca, T. Koleck, P. Borderies,

C. Albinet, L. Villard, A. Hamadi, and T. Le Toan, “Ground-based

array for tomographic imaging of the tropical forest in P-band,” IEEE

Transactions on Geoscience and Remote Sensing, vol. 51, no. 8, pp.

4460–4472, 2013.

[38] A. R. Monteith, L. M. H. Ulander, and S. Tebaldini, “Calibration of a

ground-based array radar for tomographic imaging of natural media,”

Remote Sensing, vol. 11, no. 24, pp. 1–22, 2019.

[39] J. A. Goff, “Low-pressure properties of water-from 160 to 212◦F.”

Transactions of the American Society of Heating and Ventilating En-

gineers, vol. 52, pp. 95–121, 1946.

[40] V. Cazcarra-Bes, M. Pardini, M. Tello, and K. P. Papathanassiou,

“Comparison of tomographic SAR reﬂectivity reconstruction algorithms

for forest applications at L-band,” IEEE Transactions on Geoscience and

Remote Sensing, vol. 58, no. 1, pp. 147–164, 2020.

[41] F. T. Ulaby, D. G. Long, W. J. Blackwell, C. Elachi, A. K. Fung, C. Ruf,

K. Sarabandi, H. A. Zebker, and J. Van Zyl, Microwave radar and

radiometric remote sensing. University of Michigan Press Ann Arbor,

2014.

[42] R. W. Schafer, “What is a Savitzky-Golay Filter? [lecture notes],” IEEE

Signal Processing Magazine, vol. 28, no. 4, pp. 111–117, 2011.

[43] L. M. H. Ulander, A. R. Monteith, P. Fr ¨

olind, A. Gustavsson,

A. Haglund, R. Ragnarsson, and G. Stenstr¨

om, “Airborne SAR for cali-

bration of P-band tower radar,” IGARSS 2019 - 2019 IEEE International

Geoscience and Remote Sensing Symposium, pp. 541–544, 2019.

[44] J. Lee and E. Pottier, Polarimetric Radar Imaging. Taylor & Francis

Group, Boca Raton, USA, 2009.

[45] J. Way, J. Paris, E. Kasischke, C. Slaughter, L. Viereck, N. Christensen,

M. C. Dobson, F. T. Ulaby, J. Richards, A. Milne, A. Sieber, F. J.

Ahern, D. Simonett, R. Hoffer, M. Imhoff, and J. Weber, “The effect

of changing environmental conditions on microwave signatures of forest

ecosystems: Preliminary results of the March 1988 Alaskan aircraft SAR

experiment,” International Journal of Remote Sensing, vol. 11, no. 7, pp.

1119–1144, 1990.

[46] A. Mavrovic, A. Roy, A. Royer, B. Filali, F. Boone, C. Pappas, and

O. Sonnentag, “Dielectric characterization of vegetation at L band using

an open-ended coaxial probe,” Geoscientiﬁc Instrumentation, Methods

and Data Systems, vol. 7, no. 3, pp. 195–208, 2018.

[47] M. A. El-Rayes and F. T. Ulaby, “Microwave dielectric spectrum of

vegetation-part I: Experimental observations,” IEEE Transactions on

Geoscience and Remote Sensing, vol. GE-25, no. 5, pp. 541–549, 1987.

[48] H. H ¨

anninen, Boreal and Temperate Trees in a Changing Climate:

Modelling the Ecophysiology of Seasonality. Springer Science &

Business Media, Dordrech, 2016, vol. 3.

[49] C. M¨

atzler and U. Wegm¨

uller, “Dielectric properties of freshwater ice at

microwave frequencies,” Journal of Physics D: Applied Physics, vol. 20,

no. 12, pp. 1623–1630, 1987.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/

Journal of Selected Topics in Applied Earth Observations and Remote Sensing

[50] G. Hufford, “A model for the complex permittivity of ice at frequencies

below 1 THz,” International Journal of Infrared and Millimeter Waves,

vol. 12, no. 7, pp. 677–682, 1991.

[51] J. P. Sparks, G. S. Campbell, and A. R. Black, “Water content, hydraulic

conductivity, and ice formation in winter stems of pinus contorta: A TDR

case study,” Oecologia, vol. 127, no. 4, pp. 468–475, 2001.

[52] G. I. Torgovnikov, Dielectric properties of wood and wood-based

materials. Springer-Verlag, 1993.

[53] R. Zweifel and R. H¨

asler, “Frost-induced reversible shrinkage of bark of

mature sub alpine conifers,” Agricultural and Forest Meteorology, vol.

102, no. 4, pp. 213–222, 2000.

[54] U. Wegm¨

uller, “The effect of freezing and thawing on the microwave

signatures of bare soil,” Remote Sensing of Environment, vol. 33, no. 2,

pp. 123–135, 1990.

[55] G. Bonan, Climate Change and Terrestrial Ecosystem Modeling. Cam-

bridge University Press, Cambridge, UK, 2019.

[56] J. C. Forbes and R. D. Watson, Plants in Agriculture. Cambridge

University Press, Cambridge, 1996.

[57] J. ˇ

Cerm´

ak, J. Kuˇ

cera, W. L. Bauerle, N. Phillips, and T. M. Hinckley,

“Tree water storage and its diurnal dynamics related to sap ﬂow

and changes in stem volume in old-growth Douglas-ﬁr trees,” Tree

physiology, vol. 27, no. 2, pp. 181–198, 2007.

[58] K. C. McDonald, R. Zimmermann, and J. S. Kimball, “Diurnal and

spatial variation of xylem dielectric constant in Norway spruce (Picea

abies [L.] Karst.) as related to microclimate, xylem sap ﬂow, and xylem

chemistry,” IEEE Transactions on Geoscience and Remote Sensing,

vol. 40, no. 9, pp. 2063–2082, 2002.

[59] M. Watanabe, T. Motohka, T. Shiraishi, R. B. Thapa, C. Yonezawa,

K. Nakamura, and M. Shimada, “Multitemporal ﬂuctuations in L-band

backscatter from a Japanese forest,” IEEE Transactions on Geoscience

and Remote Sensing, vol. 53, no. 11, pp. 5799–5813, 2015.

[60] T. Chiu and K. Sarabandi, “Electromagnetic scattering interaction be-

tween a dielectric cylinder and a slightly rough surface,” IEEE Transac-

tions on Antennas and Propagation, vol. 47, no. 5, pp. 902–913, 1999.

[61] G. Toole and S. Tool, New Understanding Biology for Advanced Level,

4th ed. Nelson Thornes Ltd, Cheltenham, 1999.

[62] F. A. Daudet, X. Le Roux, H. Sinoquet, and B. Adam, “Wind speed

and leaf boundary layer conductance variation within tree crown conse-

quences on leaf-to-atmosphere coupling and tree functions,” Agricultural

and Forest Meteorology, vol. 97, no. 3, pp. 171–185, 1999.

[63] C. R. Chu, C.-I. Hsieh, S.-Y. Wu, and N. G. Phillips, “Transient response

of sap ﬂow to wind speed,” Journal of Experimental Botany, vol. 60,

no. 1, pp. 249–255, 2009.

[64] M. Schlund, K. Scipal, and S. Quegan, “Assessment of a power law

relationship between P-band SAR backscatter and aboveground biomass

and its implications for BIOMASS mission performance,” IEEE Journal

of Selected Topics in Applied Earth Observations and Remote Sensing,

vol. 11, no. 10, pp. 3538–3547, 2018.

[65] S. Quegan and Jiong Jiong Yu, “Filtering of multichannel SAR images,”

IEEE Transactions on Geoscience and Remote Sensing, vol. 39, no. 11,

pp. 2373–2379, 2001.

[66] N. G. Phillips, R. Oren, J. Licata, and S. Linder, “Time series diagnosis

of tree hydraulic characteristics,” Tree Physiology, vol. 24, no. 8, pp.

879–890, 2004.

[67] F. C. Meinzer, S. A. James, G. Goldstein, and D. Woodruff, “Whole-

tree water transport scales with sapwood capacitance in tropical forest

canopy trees,” Plant, Cell and Environment, vol. 26, no. 7, pp. 1147–

1155, 2003.

[68] R. Gall, W. Landolt, P. Schleppi, V. Michellod, and J. Bucher, “Water

content and bark thickness of Norway spruce (Picea abies) stems:

phloem water capacitance and xylem sap ﬂow,” Tree Physiology, vol. 22,

no. 9, pp. 613–623, 2002.

Albert R. Monteith (S’09-M’20) received the M.Sc.

degree in electrical engineering and Ph.D. degree in

radio and space science from Chalmers University

of Technology, Gothenburg, Sweden in 2015 and

2020, respectively. He is currently a postdoctoral

researcher at the Department of Space, Earth and

Environment. His main research interests lie in the

temporal aspects of radar remote sensing of forests.

Lars M. H. Ulander (S’86-M’90-SM’04-F’17) re-

ceived the M.Sc. degree in engineering physics and

the Ph.D. degree in electrical and computer engi-

neering from Chalmers University of Technology,

Gothenburg, Sweden, in 1985 and 1991, respec-

tively. Since 1995, he has been with the Swedish

Defence Research Agency (FOI), where he is the

Director of Research in Radar Signal Processing and

leads the research on very high frequency/ultrahigh

frequency band radar. Since 2014, he has been a

Professor in radar remote sensing with Chalmers

University of Technology. He is the author or coauthor of over 300 pro-

fessional publications, of which more than 60 are in peer-reviewed scientiﬁc

journals. He is the holder of ﬁve patents. His research interests include syn-

thetic aperture radar, electromagnetic scattering models, and remote sensing

applications.

Sensitivity of P- and L-Band SAR Tomography to Above-Ground Biomass in a Hilly Temperate Forest

Article

Full-text available

Jan 2024

Tomographic Synthetic Aperture Radar (TomoSAR) is a promising technique for estimation of forest Above-Ground Biomass (AGB), but knowledge gaps still remain concerning the effects of forest type and ground topography. The paper presents new results at P- and L-band based on data acquired during the TomoSense campaign. The study area is a temperate forest, predominantly beech and spruce, with ground slopes ranging up to 40°. Analysis of vertical reflectivity profiles show distinct differences for spruce and beech. Three AGB retrieval methods are analysed, i.e. total vertical backscatter I_tot , canopy backscatter from a height layer I_c , and the ratio I_cr = I_c/I_tot . All three methods show sensitivity to AGB for spruce, whereas for beech this is only true for the two latter methods. For P-band, a significant ground slope effect is observed, while less so for L-band. The highest R ² is obtained for spruce with HV-polarisation, I_c and ground slopes less than 10°, i.e. R ² = 0.86 and RMSE = 15.6% for P-band and R ² = 0.75 and RMSE = 12.5% for L-band. Corresponding results by including all forest types are R ² = 0.77 and RMSE = 11.4% for P-band and R ² = 0.54 and RMSE = 12.0% for L-band. Moreover, performance of using I_cr is similar to that of I_c . The ratio I_cr can be determined without absolute radiometric calibration which relaxes system requirements. This paper reinforces the potential of TomoSAR for forest AGB estimation and draws attention to important effects of tree species and ground slope.

Setup of a Drone-Based SAR Experiment to Analyze a Boreal Forest

Conference Paper

Jul 2024

Remote Sensing of Sustainability

Chapter

Apr 2025

Ashok K. Singh

Remote sensing using satellites, manned or unmanned aerial vehicles (AVs), and weather balloons provides critical data for sustainability determination. This chapter introduces remote sensing devices, different sensor types, and data extraction and processing. The advantages and disadvantages of different procedures are also discussed. One of the critical disadvantages of RS is that, for developing countries, obtaining RS data for research and development purposes may be complex, thus not efficient to use RS technology to support sustainable development.

Tomographic SAR Reconstruction for Forest Height Estimation

Preprint

Full-text available

Dec 2024

Tree height estimation serves as an important proxy for biomass estimation in ecological and forestry applications. While traditional methods such as photogrammetry and Light Detection and Ranging (LiDAR) offer accurate height measurements, their application on a global scale is often cost-prohibitive and logistically challenging. In contrast, remote sensing techniques, particularly 3D tomographic reconstruction from Synthetic Aperture Radar (SAR) imagery, provide a scalable solution for global height estimation. SAR images have been used in earth observation contexts due to their ability to work in all weathers, unobscured by clouds. In this study, we use deep learning to estimate forest canopy height directly from 2D Single Look Complex (SLC) images, a derivative of SAR. Our method attempts to bypass traditional tomographic signal processing, potentially reducing latency from SAR capture to end product. We also quantify the impact of varying numbers of SLC images on height estimation accuracy, aiming to inform future satellite operations and optimize data collection strategies. Compared to full tomographic processing combined with deep learning, our minimal method (partial processing + deep learning) falls short, with an error 16-21\% higher, highlighting the continuing relevance of geometric signal processing.

Borealscat-2: Backscatter Measurements of Forest Water Dynamics

Conference Paper

Jul 2024

Uso das Geotecnologias na Estimativa de Biomassa e Carbono Florestal: Uma Revisão

Article

Full-text available

Mar 2024

As florestas tropicais são essenciais para a mitigação das mudanças climáticas, pois armazenam grandes quantidades de carbono na biomassa e na matéria orgânica do solo. Estudos utilizando dados de inventários florestais em modelos de interpolação geoestatística (krigagem) e sensoriamento remoto (imagens ópticas e radar de abertura sintética - SAR) têm revelado precisão satisfatória na estimativa da biomassa e carbono florestal. Frente à importância dessas ferramentas, o presente estudo teve como objetivo apresentar uma revisão que discorre sobre as aplicações das geotecnologias na estimativa de biomassa e carbono florestal. Estimativas mais precisas da biomassa e carbono florestal podem ser alcançadas quando a variabilidade espacial é levada em consideração. Fatores ambientais, como solo, relevo e clima, podem ser usados como preditores da distribuição espacial da biomassa e carbono florestal quando há uma relação entre essas variáveis e os diferentes tipos de vegetação, com o auxílio das geotecnologias. A krigagem, imagens de ópticas e de radar são geotecnologias que vêm auxiliando na modelagem de biomassa e carbono florestal, na construção de mapas de distribuição espacial, que podem ser utilizados como ferramentas para a orientação de políticas públicas, para inventários de emissões e estimativas de balanço de carbono, para determinar se o local atua como fonte ou sumidouro de carbono.

The influence of surface canopy water on L-band backscatter from corn: A study combining detailed In situ data and the Tor Vergata radiative transfer model

Article

Jun 2024

Detection of soil and canopy freeze/thaw state in the boreal region with L and C Band Synthetic Aperture Radar

Article

May 2024
REMOTE SENS ENVIRON

Development and Initial Evaluation of the Tower-based Borealscat-2 P- and L-Band Tomographic SAR

Conference Paper

Jul 2023

Diurnal Cycles of L-Band Tomographic SAR Backscatter in a Boreal Forest During Summer: Observations by the Borealscat Tower Radar

Conference Paper

Jul 2023

Mapping above-ground biomass in tropical forests with ground-cancelled P-band SAR and limited reference data

Article

Full-text available

Nov 2020
REMOTE SENS ENVIRON

This paper introduces the CASINO (CAnopy backscatter estimation, Subsampling, and Inhibited Nonlinear Optimisation) algorithm for above-ground biomass (AGB) estimation in tropical forests using P-band (435 MHz) synthetic aperture radar (SAR) data. The algorithm has been implemented in a prototype processor for European Space Agency's (ESA's) 7th Earth Explorer Mission BIOMASS, scheduled for launch in 2023. CASINO employs an interferometric ground cancellation technique to estimate canopy backscatter (CB) intensity. A power law model (PLM) is then used to model the dependence of CB on AGB for a large number of systematically distributed SAR data samples and a small number of calibration areas with a known AGB. The PLM parameters and AGB for the samples are estimated simultaneously within pre-defined intervals using nonlinear minimisation of a cost function. The performance of CASINO is assessed over six tropical forest sites on two continents: two in French Guiana, South America and four in Gabon, Africa, using SAR data acquired during airborne ESA campaigns and processed to simulate BIOMASS acquisitions. Multiple tests with only two randomly selected calibration areas with AGB > 100 t/ha are conducted to assess AGB estimation performance given limited reference data. At 2.25 ha scale and using a single flight heading, the root-mean-square difference (RMSD) is ≤ 27% for at least 50% of all tests in each test site and using as reference AGB maps derived from airborne laser scanning data. An improvement is observed when two flight headings are used in combination. The most consistent AGB estimation (lowest RMSD variation across different calibration sets) is observed for test sites with a large AGB interval and average AGB around 200–250 t/ha. The most challenging conditions are in areas with AGB < 200 t/ha and large topographic variations. A comparison with 142 1 ha plots distributed across all six test sites and with AGB estimated from in situ measurements gives an RMSD of 20% (66 t/ha).

The BIOMASS Level 2 Prototype Processor: Design and Experimental Results of Above-Ground Biomass Estimation

Article

Full-text available

Mar 2020

BIOMASS is ESA’s seventh Earth Explorer mission, scheduled for launch in 2022. The satellite will be the first P-band SAR sensor in space and will be operated in fully polarimetric interferometric and tomographic modes. The mission aim is to map forest above-ground biomass (AGB), forest height (FH) and severe forest disturbance (FD) globally with a particular focus on tropical forests. This paper presents the algorithms developed to estimate these biophysical parameters from the BIOMASS level 1 SAR measurements and their implementation in the BIOMASS level 2 prototype processor with a focus on the AGB product. The AGB product retrieval uses a physically-based inversion model, using ground-canceled level 1 data as input. The FH product retrieval applies a classical PolInSAR inversion, based on the Random Volume over Ground Model (RVOG). The FD product will provide an indication of where significant changes occurred within the forest, based on the statistical properties of SAR data. We test the AGB retrieval using modified airborne P-Band data from the AfriSAR and TropiSAR campaigns together with reference data from LiDAR-based AGB maps and plot-based ground measurements. For AGB estimation based on data from a single heading, comparison with reference data yields relative Root Mean Square Difference (RMSD) values mostly between 20% and 30%. Combining different headings in the estimation process significantly improves the AGB retrieval to slightly less than 20%. The experimental results indicate that the implemented retrieval scheme provides robust results that are within mission requirements.

Calibration of a Ground-Based Array Radar for Tomographic Imaging of Natural Media

Article

Full-text available

Dec 2019

Ground-based tomographic radar measurements provide valuable knowledge about the electromagnetic scattering mechanisms and temporal variations of an observed scene and are essential in preparation for space-borne tomographic synthetic aperture radar (SAR) missions. Due to the short range between the radar antennas and a scene being observed, the tomographic radar observations are affected by several systematic errors. This article deals with the modelling and calibration of three systematic errors: mutual antenna coupling, magnitude and phase errors and the pixel-variant impulse response of the tomographic image. These errors must be compensated for so that the tomographic images represent an undistorted rendering of the scene reflectivity. New calibration methods were described, modelled and validated using experimental data. The proposed methods will be useful for future ground-based tomographic radar experiments in preparation for space-borne SAR missions.

Airborne SAR for Calibration of P-Band Tower Radar

Conference Paper

Jul 2019

Comparison of Tomographic SAR Reflectivity Reconstruction Algorithms for Forest Applications at L-band

Article

Sep 2019

Forest structure is a key parameter for forest applications, but it is difficult to be estimated at the required spatial and temporal scales. In this context, synthetic aperture radar Tomography (TomoSAR) that allows, at lower frequencies, the 3-D imaging of natural volume scatterers with high spatial and temporal resolution may be a game changer. The aim of this article is to evaluate three TomoSAR algorithms, Fourier beamforming (FB), Capon beamforming (CB), and compressive sensing (CS) with respect to their performance in the reconstruction of the 3-D forest reflectivity. The implications of volumetric forest scattering, as well as the temporal decorrelation of scatterers, are analyzed. The algorithms are compared on a set of simulated scenarios and then evaluated on an experimental L-band data set composed by four acquisition dates, each one consisting of five tomographic tracks. The data were acquired in 2014, within a time span of two months, over the Traunstein forest (Germany) using the F-SAR system. Additionally, discrete airborne Lidar has been used for a qualitative evaluation. The results indicate that the CS reconstruction is, for many practical cases, superior when compared to FB or CB reconstructions as they achieve higher vertical resolution, especially in cases with a lower number of acquisitions and complex forest scenarios. By combining acquisitions performed at different days, the effect of temporal decorrelation on each algorithm for two different tomographic implementations (repeat-pass vs. single-pass) has been assessed. The results indicate that simultaneously acquired image pairs allow a better reconstruction of the 3-D forest reflectivity.

Soil and Vegetation Scattering Contributions in L-Band and P-Band Polarimetric SAR Observations

Article

Jun 2019

Active microwave-based retrieval of soil moisture in vegetated areas has uncertainties due to the sensitivity of the signal to both soil (dielectric constant and roughness) and vegetation (dielectric constant and structure) properties. A multi-frequency acquisition system would increase the number of observations that may constrain soil and/or vegetation parameter retrievals. In order to realize this constraint, an understanding of microwaves interaction with the surface and vegetation across frequencies is necessary. Different microwave frequencies have varied interactions with the soil-vegetation medium and increasing penetration into the soil and canopy with the decreasing frequency. In this study, we examine the contributions of different scattering mechanisms to coincident observations from two microwave frequencies (L and P) of airborne synthetic aperture radar instruments. We quantify contributions of surface, vegetation volume, and double-bounce scattering components. Results are analyzed and discussed to guide future multi-frequency retrieval algorithm designs.

The European Space Agency BIOMASS mission: Measuring forest above-ground biomass from space

Article